This thesis describes a question answering system, which takes advantages of Genetic Algorithms for extracting answers from web snippets. These GA learn the syntactic alignment between pairs fsentence, answerg obtained from past QA cycles in order to identify and extract the most promising answers to new natural language questions. The answer extraction strategy is strongly data-driven using only language specific top-lists and thus, the whole approach has a high degree of language independency. In this thesis, ideas of how to add linguistic processing to this data-driven search are also discussed. The strategies were assessed with different sets of pairs question, answer. Results show that this approach is promising, especially, when it deals with specific questions.